Interruption costs are the full cognitive, temporal, and quality expenses incurred when a knowledge worker's sustained engagement is disrupted. The costs include immediate effects (the time spent handling the interruption itself), recovery lag (the 23+ minutes Gloria Mark measured before performance returns to pre-interruption levels), attention residue (the persistence of interrupted-task elements degrading subsequent work), and constellation destruction (the loss of the assembled cognitive state that sustained engagement had built). For complex knowledge work, especially AI-augmented creative tasks that produce deep flow states, the total cost of a single interruption can exceed an hour of degraded performance even when the interruption itself lasted seconds. Organizations treating interruptions as brief, isolated events systematically underestimate their consequences and design workflows that maximize rather than minimize these compounding costs.
The compound structure is what conventional analysis misses. An interruption that takes two minutes to handle — a colleague's question, a notification requiring response, an agent's output needing quick evaluation — appears to cost two minutes. The actual cost includes the 23-minute recovery lag that follows, the residue from the interruption that degrades the next hour of work, and the destruction of the flow state that took thirty minutes to build. The total easily exceeds an hour of impaired cognitive function per two-minute interruption. The ratio — 60:2 — reveals why interruption management is not a minor optimization but a primary determinant of knowledge-worker effectiveness.
Research distinguishes between self-interruptions (internally generated switches) and external interruptions (imposed by others or by systems). Gloria Mark's naturalistic studies found that most interruptions are self-generated: the worker checks email, opens a new browser tab, decides to handle a quick administrative task between coding sessions. The self-generated pattern suggests that interruptions are not primarily an external imposition but a response to work environments that make switching feel necessary or rewarding. AI tools intensify the pattern by creating more legitimate reasons to switch: when an agent completes a task, checking its output feels productive; when a question arises that AI could answer, asking feels efficient. Each feels like a small, rational choice; the aggregate produces the chronic fragmentation that residue research demonstrates is cognitively ruinous.
The destroyed flow is arguably the highest cost, though the hardest to quantify. Mihaly Csikszentmihalyi's research established that flow states produce not just subjective satisfaction but objectively better work — more creative solutions, fewer errors, deeper insights. Assembling the conditions for flow requires time and uninterrupted engagement; a single interruption can shatter the state, requiring substantial reinvestment to re-achieve it. For AI-augmented work that is intrinsically flow-conducive (tight feedback loops, immediate results, matched challenge-skill), interruptions destroy the very state the tools were designed to enable. The builder is left with a tool capable of producing flow and a work structure that prevents it.
Organizational countermeasures exist and are specific. Interrupt batching consolidates the interruptions a builder must handle into defined periods, preserving extended focus windows between batches. Do-not-disturb protocols give builders authority to defer interruptions during deep work sessions. Asynchronous communication reduces the expectation of immediate response, converting interruptions into messages that can be handled during designated switching periods. Role specialization assigns monitoring to dedicated evaluators, allowing others to maintain sustained engagement without interruption. Each countermeasure reduces total switching frequency and protects the cognitive resources on which quality depends. The scarcity of organizations implementing these measures reveals a gap between what the research demonstrates and what the organizational culture is willing to restructure.
Interruption research has a long history in ergonomics, human factors, and cognitive psychology, with foundational work dating to studies of operator attention in safety-critical environments (aviation, nuclear control). Gloria Mark's 2000s studies brought rigorous measurement to knowledge work contexts, quantifying costs that practitioners had long felt but couldn't specify. The synthesis with attention residue research (Leroy), flow interruption consequences (Csikszentmihalyi), and the AI-era intensification represents the current frontier, where decades of cognitive science converge on a single prescription: minimize interruptions, not as a personal preference but as an engineering requirement for preserving the judgment quality on which organizational performance depends.
Compound, not isolated. Interruption costs include handling time, recovery lag, residue generation, and flow destruction — components that sum to expenses dramatically exceeding the interruption's duration.
Self-generated majority. Most knowledge-work interruptions are internally initiated, suggesting that work environments make switching feel necessary or rewarding and that individual discipline is insufficient without structural change.
Flow destruction premium. Interrupting flow-state work destroys the assembled cognitive constellation that took 20–30 minutes to build, requiring reinvestment of assembly time before deep work can resume.
Organizational design leverage. Countermeasures exist (batching, protocols, asynchronous communication, role specialization) and are actionable, but adoption remains scarce despite robust supporting evidence.